In this paper we propose to use entropy as a measure of risk. We describe the framework of our approach towards risk assessment, which can be said to be information theoretical in some sense. We present different criteria as different systems and different values of criteria as different states of these systems. Probabilities of a system being in particular states are interval-valued. In our paper we present the problem of entropy generalization to the interval-valued case and we show some of the possible solutions, viz., ways of generalization.